An unexpected campaign result should begin an investigation, not an immediate explanation. A change in conversion can originate in customer behaviour, campaign composition, tracking coverage, delayed outcomes or a revised definition.
This example workflow follows a campaign performance question from initial signal to documented action. It combines Covorg’s marketing analysis, reporting and collaboration capabilities within a defined process that makes source reconciliation and uncertainty visible.
Capture the observation precisely
Record the affected measure, campaign population, period, filters and comparison. State whether the movement is a difference in counts, a relative percentage change or a percentage point change in a rate. These forms communicate different magnitudes and should not be substituted for one another.
Preserve the definition and source state at the time the question was raised. A live dashboard may change as conversions mature or delayed events arrive. Keeping an observation reference allows the team to explain why the initial signal differed from a later result.
Rule out measurement changes first
Check spend completeness, event collection, campaign identifiers, deduplication and time zone alignment. Review whether a tracking change or source permission issue altered the observed population. Comparing a complete previous period with an incomplete current period can create a false performance decline.
| Check | Evidence to inspect | Possible interpretation |
|---|---|---|
| Spend arrival | Expected source deliveries and control totals | A partial input understates cost |
| Event coverage | Event counts and schema validation results | Collection changed rather than demand |
| Cohort maturity | Time elapsed since each enquiry | Outcomes are still developing |
| Identity mapping | Duplicate and unmatched records | Population changed after resolution |
| Attribution rule | Window and model version | Credit allocation changed |
Decompose the movement with a stable denominator
Examine channel, audience, device or location only where the available data supports the grouping. A change in overall conversion can result from a different mix of groups even when each group’s own conversion is unchanged. Review both composition and within group performance.
The example query below counts enquiries and outcomes from a prepared cohort table. It assumes one row per eligible enquiry and an agreed outcome window. Calculate rates from these aggregate counts, and handle a zero denominator explicitly rather than reporting a misleading result.
SELECT
channel,
COUNT(*) AS eligible_enquiries,
SUM(CASE WHEN converted THEN 1 ELSE 0 END)
AS converted_enquiries
FROM approved_campaign_cohort
WHERE cohort_period = :cohort_period
GROUP BY channel;Document the explanation and the next action
Separate confirmed data issues, measured business changes and hypotheses requiring further investigation. Attribution analysis can explain how credit was assigned under a rule; it does not independently prove the incremental effect of a campaign.
Agree an action owner, an evaluation period and the evidence that will determine whether the action was useful. That action might be repairing event collection, reviewing audience composition or designing an experiment. Retain the original question and reviewed evidence so subsequent reporting reflects what the team learned.
Example campaign investigation sequence
- 01
Observe
Capture the metric movement and its exact comparison context.
- 02
Validate
Check source completeness, cohort maturity and definition changes.
- 03
Decompose
Compare segment composition and performance using stable populations.
- 04
Review
Separate measured facts from causal hypotheses and missing evidence.
- 05
Follow through
Assign an action and evaluate the outcome over an agreed period.
Design considerations
- Avoid combining channel rates by taking a simple average.
- Keep incomplete customer journeys visible when evaluating recent cohorts.
- Protect customer level detail when sharing campaign investigations.
Sources
The following sources explain the underlying technical patterns.
